Fix plan unparsing for filters with subqueries involved type coercion

[sgrebnov]

DataFusion optimization uses aliases to preserve original expression names during optimizations and type coercions:
apache/datafusion#3794

Datafusion contains special logic to prevent aliasing for filter/join plans, but I still see the alias can be produced for some complex queries/cases
https://github.com/apache/datafusion/blob/f044bc8371d5b4e1e51a9026f3eccac16a6d4648/datafusion/optimizer/src/utils.rs#L316

Self {
            use_alias: !matches!(plan, LogicalPlan::Filter(_) | LogicalPlan::Join(_)),
        }

If there is a filter with additional alias information, we must use the expression only as original expression name is unnecessary and the alias can't be part of the WHERE clause. Note: we don't currently support Expr::Alias as general unparsing mechanism, put fix directly to where_expr so we are sure that the Alias is part of filter / where clause.

Example query

select c_custkey
from customer
where c_custkey > (select 1 from customer where c_acctbal > 0.00)

There will be the alias with name "customer.c_acctbal > Decimal128(Some(0),38,15)" produced due to type coercion Decimal128(Some(0),38,2) and Decimal128(Some(0),38,15)

/// Ensure `expr` has the name as `original_name` by adding an
    /// alias if necessary.
    pub fn alias_if_changed(self, original_name: String) -> Result<Expr> {
        let new_name = self.name_for_alias()?;

        if new_name == original_name { # customer.c_acctbal > Decimal128(Some(0),38,15) != customer.c_acctbal > Decimal128(Some(0),38, 2)
            return Ok(self);
        }

        Ok(self.alias(original_name))
    }

alias:Alias { expr: BinaryExpr(BinaryExpr { left: Column(Column { relation: None, name: "c_acctbal" }), op: Gt, right: Literal(Decimal128(Some(0),38,2)) }), relation: None, name: "customer.c_acctbal > Decimal128(Some(0),38,15)" }